Bellabeat Data Analysis Case Study – Google Data Analytics Certification Capstone Project

2026-02-14

Dylan Roche

I. Introduction / Background

Founded in 2014 by Urška Sršen and Sandro Mur, Bellabeat is a health and fitness technology company that primarily develops smart wellness products for women. Some of its key products include the following:

These devices connect to a dedicated Bellabeat app from which users can view activity, sleep, stress, and menstrual cycle data and make informed decisions about their health.

Bellabeat has offices located in North America, Europe, and Asia and relies mainly on digital marketing as its advertising strategy. Specifically, it has invested heavily in Google Search, maintains active social media pages that drive consumer engagement, and runs video ad campaigns on YouTube.

Even though Bellabeat has already seen substantial success as a small company, it desires to become a larger player in the wellness and technology space. Therefore, Sršen believes that an inquiry into available consumer data would unveil key insights and unlock more opportunities for growth.

To support this goal, I have assumed the role of a junior data analyst and will conduct a comprehensive analysis on user activity data. The insights obtained from this analysis will be leveraged to develop high-level recommendations for Bellabeat’s marketing strategy.

My data analysis will be documented across the following phases:

II. The Ask Phase

During the Ask phase, the business task must be determined.

Business Task

Using available daily user activity data, determine key trends in smart device usage to identify when users are least and most active. Furthermore, scrutinize potential relationships between different kinds of activity, as well as correlations between activity and sleep data. These insights will be applied to Bellabeat’s Leaf–a wellness tracker than can be worn as a bracelet, necklace, or clip–to better understand user behavior and influence product and marketing decisions.

III. The Prepare Phase

During the Prepare phase, the data is gathered and evaluated to ensure its accuracy, credibility, and integrity.

The FitBit Fitness Tracker Data dataset, initially published on Zenodo and subsequently made available through Mobius on Kaggle, is used for this analysis.

Dataset Information

The dataset consists of personal tracker data from approximately 30 FitBit users. In particular, it includes minute-level measurements for physical activity, sleep monitoring, and heart rate, as well as daily summaries of activity, step counts, calories burned, and other comparable metrics. For this analysis, to ensure that the data is consistent is possible and streamlined, I’ve decided to work strictly with daily-level data over a one-month period.

The information was collected through Amazon Mechnical Turk (MTurk)–a crowdsourcing marketplace that allows individuals to remotely complete tasks. As part of the study, participants voluntarily submitted their personal FitBit data.

Data was collected over a one-month period, from April 12, 2016 to May 12, 2016. Even though the dataset is not recent, user activity patterns related to daily movement and sleep behavior are not expected to significantly deviate if the study was conducted today, making the dataset acceptable for analysis.

Dataset Issues

While the sample of size of approximately 30 users is sufficient for this analysis, it is not large enough to properly represent the broader smart device user population. This therefore introduces selection bias, as the participants are not indicative of all This is known as selection bias.

Since the data was voluntarily submitted by users, the dataset is also subject to self-selection bias. Individuals who took part in the study are more likely to be health-conscious, physically and technologically savvy than the average consumer. Moreover, given that the data is derived only from FitBit users, it may not reflect usage patterns affiliated with other wearable devices.

Dataset Limitations

Participant-level information remains limited, as data on age, gender, and location was not disclosed. This restricts the ability segment users and analyze behavioral differences across demographic groups.

Some users did not consistently enter activity on specific days, and not all metrics were recorded uniformly across all participants. These inconsistencies could potentially impact trend analysis and restrict the accuracy of comparisons across users.

IV. The Process Phase

During the Process phase, the tools used for analysis are determined, and the data is cleaned.

Analysis Tools

There are a variety of tools that can be utilized for data analysis,including Microsoft Excel, SQL, Python, and R. Often, a data analysis workflow spans numerous programs, with each tool serving a unique purpose.For this analysis, Microsoft Excel was used for initial data cleaning and basic feature engineering, while R was selected for more complex dataset preprocessing, feature engineering, and dataset transformation tasks.

Initial Data Cleaning and Feature Engineering in Microsoft Excel

To begin, the Daily Activity (Merged) CSV file opened in Microsoft Excel to gain an initial understanding of the dataset–specifically, the columns, structure, and overall distribution. The data was presented in a wide format, meaning that multiple related metrics are distributed across different columns. While this format does not pose any immediate challenges for analysis, subsets of the data were later converted to long format to facilitate data visualization.

Excel’s built-in Remove Duplicates feature was used to check duplicates entries; none were found. Next, missing values were assessed using the Find & Select feature to identify blank cells, and no missing data was detected. The data column was then reformatted to YYYY-MM-DD, ensuring consistency with R’s date format and enabling a seamless transition between tools.

Understanding the type of day–whether a weekday or weekend–can prove particularly useful when analyzing activity patterns. Thus, an IsWeekend column was added to the dataset by writing the formula: =WEEKDAY(B2, 2) \> 5, which returns TRUE for weekend dates and FALSE for weekdays.

Data Processing and Feature Engineering in R

First, the required libraries were installed and loaded into R. Next, the Daily Activity, Daily Sleep, and Weight Log datasets were imported.

# Loading in the required packages 
library(tidyverse)
library(dplyr)
library(ggplot2)
library(ggpubr)
library(ggcorrplot)
# Importing the "Daily Activity" dataset
daily_activity <- read.csv("C:/Users/melfl/Downloads/archive/mturkfitbit_export_4.12.16-5.12.16/Fitabase Data 4.12.16-5.12.16/DailyActivityDataset.csv")
head(daily_activity)
glimpse(daily_activity)

# Importing the "Daily Sleep" dataset 
daily_sleep <- read.csv("C:/Users/melfl/Downloads/archive (1)/mturkfitbit_export_4.12.16-5.12.16/Fitabase Data 4.12.16-5.12.16/DailySleepDataset.csv")
head(daily_sleep)
glimpse(daily_sleep)

# Importing the "Weight Log" dataset
weight_log <- read.csv("C:/Users/melfl/Downloads/archive/mturkfitbit_export_4.12.16-5.12.16/Fitabase Data 4.12.16-5.12.16/weightLogInfo_merged.csv")
head(weight_log)
glimpse(weight_log)

activity_df = read.csv("C:/Users/melfl/Downloads/archive/mturkfitbit_export_4.12.16-5.12.16/Fitabase Data 4.12.16-5.12.16/dailyActivity_merged.csv")

Once the datasets were imported, the number of participants included in each dataset was calculated.

activity_df = read.csv("C:/Users/melfl/Downloads/archive/mturkfitbit_export_4.12.16-5.12.16/Fitabase Data 4.12.16-5.12.16/dailyActivity_merged.csv")

# Finding the number of participants in each dataset
activity_participants <- n_distinct(activity_df$Id)
sleep_participants <- n_distinct(daily_sleep$Id)
weight_participants <- n_distinct(weight_log$Id)
print(activity_participants)
## [1] 33
print(sleep_participants)
## [1] 24
print(weight_participants)
## [1] 8

The results of the above calculations were then visualized via a bar chart.

In statistics, the Central Limit Theorem states that, as the sample sizes increases, the sampling distribution of the mean increasingly approaches a normal distribution. A regularly used rule of thumb is that a sample size of n ≥ 30 is acceptable for this approximation.

The Weight dataset only contains 8 participants, which is well below this guideline. Therefore, it will not be included in this analysis. The Sleep dataset consists of 24 participants, which is slightly below the recommended threshold; however, it will be kept for exploratory data analysis (EDA) and to examine potential correlations between activity and sleep patterns.

Even though the date column was reformatted in Excel earlier, R still intreprets it as a string rather than a date.

glimpse(daily_activity)
## Rows: 940
## Columns: 16
## $ Id                       <dbl> 1503960366, 1503960366, 1503960366, 150396036…
## $ ActivityDate             <chr> "2016-04-12", "2016-04-13", "2016-04-14", "20…
## $ TotalSteps               <int> 13162, 10735, 10460, 9762, 12669, 9705, 13019…
## $ TotalDistance            <dbl> 8.50, 6.97, 6.74, 6.28, 8.16, 6.48, 8.59, 9.8…
## $ TrackerDistance          <dbl> 8.50, 6.97, 6.74, 6.28, 8.16, 6.48, 8.59, 9.8…
## $ LoggedActivitiesDistance <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ VeryActiveDistance       <dbl> 1.88, 1.57, 2.44, 2.14, 2.71, 3.19, 3.25, 3.5…
## $ ModeratelyActiveDistance <dbl> 0.55, 0.69, 0.40, 1.26, 0.41, 0.78, 0.64, 1.3…
## $ LightActiveDistance      <dbl> 6.06, 4.71, 3.91, 2.83, 5.04, 2.51, 4.71, 5.0…
## $ SedentaryActiveDistance  <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ VeryActiveMinutes        <int> 25, 21, 30, 29, 36, 38, 42, 50, 28, 19, 66, 4…
## $ FairlyActiveMinutes      <int> 13, 19, 11, 34, 10, 20, 16, 31, 12, 8, 27, 21…
## $ LightlyActiveMinutes     <int> 328, 217, 181, 209, 221, 164, 233, 264, 205, …
## $ SedentaryMinutes         <int> 728, 776, 1218, 726, 773, 539, 1149, 775, 818…
## $ Calories                 <int> 1985, 1797, 1776, 1745, 1863, 1728, 1921, 203…
## $ IsWeekend                <lgl> FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, FALSE…
glimpse(daily_sleep)
## Rows: 413
## Columns: 6
## $ Id                 <dbl> 1503960366, 1503960366, 1503960366, 1503960366, 150…
## $ SleepDay           <chr> "2016-04-12", "2016-04-13", "2016-04-15", "2016-04-…
## $ TotalSleepRecords  <int> 1, 2, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
## $ TotalMinutesAsleep <int> 327, 384, 412, 340, 700, 304, 360, 325, 361, 430, 2…
## $ TotalTimeInBed     <int> 346, 407, 442, 367, 712, 320, 377, 364, 384, 449, 3…
## $ IsWeekend          <lgl> FALSE, FALSE, FALSE, TRUE, TRUE, FALSE, FALSE, FALS…

Therefore, it is necessary to convert the data variables in both datasets to the proper date type. Afterwards, columns not required for the analysis were removed.

# Converting "ActivityDate" to a date-type
daily_activity$ActivityDate_Fixed <- ymd(daily_activity$ActivityDate) 
daily_activity <- select(daily_activity, Id, -ActivityDate, ActivityDate_Fixed, 3:5, -LoggedActivitiesDistance, 7:16) 
glimpse(daily_activity) 
head(daily_activity)


# Converting "SleepDay" to a date-type 
daily_sleep$SleepDate_Fixed <- ymd(daily_sleep$SleepDay) 
daily_sleep <- select(daily_sleep, Id, -SleepDay, SleepDate_Fixed, 3:6) 
glimpse(daily_sleep) 
head(daily_sleep)

Daily Activity Dataset:

## Rows: 940
## Columns: 15
## $ Id                       <dbl> 1503960366, 1503960366, 1503960366, 150396036…
## $ ActivityDate_Fixed       <date> 2016-04-12, 2016-04-13, 2016-04-14, 2016-04-…
## $ TotalSteps               <int> 13162, 10735, 10460, 9762, 12669, 9705, 13019…
## $ TotalDistance            <dbl> 8.50, 6.97, 6.74, 6.28, 8.16, 6.48, 8.59, 9.8…
## $ TrackerDistance          <dbl> 8.50, 6.97, 6.74, 6.28, 8.16, 6.48, 8.59, 9.8…
## $ VeryActiveDistance       <dbl> 1.88, 1.57, 2.44, 2.14, 2.71, 3.19, 3.25, 3.5…
## $ ModeratelyActiveDistance <dbl> 0.55, 0.69, 0.40, 1.26, 0.41, 0.78, 0.64, 1.3…
## $ LightActiveDistance      <dbl> 6.06, 4.71, 3.91, 2.83, 5.04, 2.51, 4.71, 5.0…
## $ SedentaryActiveDistance  <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ VeryActiveMinutes        <int> 25, 21, 30, 29, 36, 38, 42, 50, 28, 19, 66, 4…
## $ FairlyActiveMinutes      <int> 13, 19, 11, 34, 10, 20, 16, 31, 12, 8, 27, 21…
## $ LightlyActiveMinutes     <int> 328, 217, 181, 209, 221, 164, 233, 264, 205, …
## $ SedentaryMinutes         <int> 728, 776, 1218, 726, 773, 539, 1149, 775, 818…
## $ Calories                 <int> 1985, 1797, 1776, 1745, 1863, 1728, 1921, 203…
## $ IsWeekend                <lgl> FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, FALSE…
##           Id ActivityDate_Fixed TotalSteps TotalDistance TrackerDistance
## 1 1503960366         2016-04-12      13162          8.50            8.50
## 2 1503960366         2016-04-13      10735          6.97            6.97
## 3 1503960366         2016-04-14      10460          6.74            6.74
## 4 1503960366         2016-04-15       9762          6.28            6.28
## 5 1503960366         2016-04-16      12669          8.16            8.16
## 6 1503960366         2016-04-17       9705          6.48            6.48
##   VeryActiveDistance ModeratelyActiveDistance LightActiveDistance
## 1               1.88                     0.55                6.06
## 2               1.57                     0.69                4.71
## 3               2.44                     0.40                3.91
## 4               2.14                     1.26                2.83
## 5               2.71                     0.41                5.04
## 6               3.19                     0.78                2.51
##   SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes
## 1                       0                25                  13
## 2                       0                21                  19
## 3                       0                30                  11
## 4                       0                29                  34
## 5                       0                36                  10
## 6                       0                38                  20
##   LightlyActiveMinutes SedentaryMinutes Calories IsWeekend
## 1                  328              728     1985     FALSE
## 2                  217              776     1797     FALSE
## 3                  181             1218     1776     FALSE
## 4                  209              726     1745     FALSE
## 5                  221              773     1863      TRUE
## 6                  164              539     1728      TRUE

Daily Sleep Dataset:

## Rows: 413
## Columns: 6
## $ Id                 <dbl> 1503960366, 1503960366, 1503960366, 1503960366, 150…
## $ SleepDate_Fixed    <date> 2016-04-12, 2016-04-13, 2016-04-15, 2016-04-16, 20…
## $ TotalSleepRecords  <int> 1, 2, 1, 2, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, …
## $ TotalMinutesAsleep <int> 327, 384, 412, 340, 700, 304, 360, 325, 361, 430, 2…
## $ TotalTimeInBed     <int> 346, 407, 442, 367, 712, 320, 377, 364, 384, 449, 3…
## $ IsWeekend          <lgl> FALSE, FALSE, FALSE, TRUE, TRUE, FALSE, FALSE, FALS…
##           Id SleepDate_Fixed TotalSleepRecords TotalMinutesAsleep
## 1 1503960366      2016-04-12                 1                327
## 2 1503960366      2016-04-13                 2                384
## 3 1503960366      2016-04-15                 1                412
## 4 1503960366      2016-04-16                 2                340
## 5 1503960366      2016-04-17                 1                700
## 6 1503960366      2016-04-19                 1                304
##   TotalTimeInBed IsWeekend
## 1            346     FALSE
## 2            407     FALSE
## 3            442     FALSE
## 4            367      TRUE
## 5            712      TRUE
## 6            320     FALSE

The number of missing and duplicate entries in both datasets were verified once more.

# Finding the number of missing values and duplicates in the "Daily Activity" dataset
sum(is.na(daily_activity))
## [1] 0
sum(is.na(daily_sleep))
## [1] 0
sum(duplicated(daily_activity))
## [1] 0
# Finding the number of and removing the missing values and duplicates in the "Daily Sleep" dataset
sum(duplicated(daily_sleep))
## [1] 3
duplicated(daily_sleep)
##   [1] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [13] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [25] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [37] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [49] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [61] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [73] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [85] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
##  [97] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [109] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [121] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [133] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [145] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [157] FALSE FALSE FALSE FALSE FALSE  TRUE FALSE FALSE FALSE FALSE FALSE FALSE
## [169] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [181] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [193] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [205] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [217] FALSE FALSE FALSE FALSE FALSE FALSE FALSE  TRUE FALSE FALSE FALSE FALSE
## [229] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [241] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [253] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [265] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [277] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [289] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [301] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [313] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [325] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [337] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [349] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [361] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [373] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE  TRUE FALSE FALSE FALSE
## [385] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [397] FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE FALSE
## [409] FALSE FALSE FALSE FALSE FALSE

The three duplicates in the Daily Sleep dataset were removed.

# Removing th missing values and duplicates in the "Daily Sleep" dataset
daily_sleep_clean <- daily_sleep[!duplicated(daily_sleep), ]
sum(duplicated(daily_sleep_clean))
## [1] 0

Now that both datasets were cleaned and checked for data integrity, they were joined utilizing a left join based on the shared identifier columns (primary keys). A left join was used to preserve all records from the Daily Activity dataset. Since the Daily Sleep contains fewer participants than the Daily Activity dataset, some activity records do not have corresponding sleep data; however, these entries were retained to avoid unnecessary data loss.

# Combining the "Daily Activity" and "Daily Sleep" datasets with a left join
fitness_data_df <- left_join(daily_activity, daily_sleep_clean, by= c("Id" = "Id", "ActivityDate_Fixed" = "SleepDate_Fixed", "IsWeekend" = "IsWeekend"))
head(fitness_data_df)
##           Id ActivityDate_Fixed TotalSteps TotalDistance TrackerDistance
## 1 1503960366         2016-04-12      13162          8.50            8.50
## 2 1503960366         2016-04-13      10735          6.97            6.97
## 3 1503960366         2016-04-14      10460          6.74            6.74
## 4 1503960366         2016-04-15       9762          6.28            6.28
## 5 1503960366         2016-04-16      12669          8.16            8.16
## 6 1503960366         2016-04-17       9705          6.48            6.48
##   VeryActiveDistance ModeratelyActiveDistance LightActiveDistance
## 1               1.88                     0.55                6.06
## 2               1.57                     0.69                4.71
## 3               2.44                     0.40                3.91
## 4               2.14                     1.26                2.83
## 5               2.71                     0.41                5.04
## 6               3.19                     0.78                2.51
##   SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes
## 1                       0                25                  13
## 2                       0                21                  19
## 3                       0                30                  11
## 4                       0                29                  34
## 5                       0                36                  10
## 6                       0                38                  20
##   LightlyActiveMinutes SedentaryMinutes Calories IsWeekend TotalSleepRecords
## 1                  328              728     1985     FALSE                 1
## 2                  217              776     1797     FALSE                 2
## 3                  181             1218     1776     FALSE                NA
## 4                  209              726     1745     FALSE                 1
## 5                  221              773     1863      TRUE                 2
## 6                  164              539     1728      TRUE                 1
##   TotalMinutesAsleep TotalTimeInBed
## 1                327            346
## 2                384            407
## 3                 NA             NA
## 4                412            442
## 5                340            367
## 6                700            712
glimpse(fitness_data_df)
## Rows: 940
## Columns: 18
## $ Id                       <dbl> 1503960366, 1503960366, 1503960366, 150396036…
## $ ActivityDate_Fixed       <date> 2016-04-12, 2016-04-13, 2016-04-14, 2016-04-…
## $ TotalSteps               <int> 13162, 10735, 10460, 9762, 12669, 9705, 13019…
## $ TotalDistance            <dbl> 8.50, 6.97, 6.74, 6.28, 8.16, 6.48, 8.59, 9.8…
## $ TrackerDistance          <dbl> 8.50, 6.97, 6.74, 6.28, 8.16, 6.48, 8.59, 9.8…
## $ VeryActiveDistance       <dbl> 1.88, 1.57, 2.44, 2.14, 2.71, 3.19, 3.25, 3.5…
## $ ModeratelyActiveDistance <dbl> 0.55, 0.69, 0.40, 1.26, 0.41, 0.78, 0.64, 1.3…
## $ LightActiveDistance      <dbl> 6.06, 4.71, 3.91, 2.83, 5.04, 2.51, 4.71, 5.0…
## $ SedentaryActiveDistance  <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ VeryActiveMinutes        <int> 25, 21, 30, 29, 36, 38, 42, 50, 28, 19, 66, 4…
## $ FairlyActiveMinutes      <int> 13, 19, 11, 34, 10, 20, 16, 31, 12, 8, 27, 21…
## $ LightlyActiveMinutes     <int> 328, 217, 181, 209, 221, 164, 233, 264, 205, …
## $ SedentaryMinutes         <int> 728, 776, 1218, 726, 773, 539, 1149, 775, 818…
## $ Calories                 <int> 1985, 1797, 1776, 1745, 1863, 1728, 1921, 203…
## $ IsWeekend                <lgl> FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, FALSE…
## $ TotalSleepRecords        <int> 1, 2, NA, 1, 2, 1, NA, 1, 1, 1, NA, 1, 1, 1, …
## $ TotalMinutesAsleep       <int> 327, 384, NA, 412, 340, 700, NA, 304, 360, 32…
## $ TotalTimeInBed           <int> 346, 407, NA, 442, 367, 712, NA, 320, 377, 36…

V. The Analyze Phase

During the Analyze phase, key trends, patterns, and relationships are determined within the data.

Data Analysis on the Daily Activity Dataset

To begin, the key variables in the Daily Activity Dataset were analyzed.

TotalSteps Analysis

Summary statistics were calculated for the TotalSteps variable–grouped by participant and type of day, respectively.

# Summary statistics for "TotalSteps" (grouped by participant) 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_total_steps = mean(TotalSteps),
    median_total_steps = median(TotalSteps), 
    sd_total_steps = sd(TotalSteps),
    min_total_steps = min(TotalSteps),
    max_total_steps = max(TotalSteps),
    iqr_total_steps = IQR(TotalSteps)
  )
## # A tibble: 33 × 7
##            Id mean_total_steps median_total_steps sd_total_steps min_total_steps
##         <dbl>            <dbl>              <dbl>          <dbl>           <int>
##  1 1503960366           12117.             12207           3052.               0
##  2 1624580081            5744.              4026           6177.            1510
##  3 1644430081            7283.              6684.          4325.            1223
##  4 1844505072            2580.              2237           2713.               0
##  5 1927972279             916.               152           1205.               0
##  6 2022484408           11371.             11548           2807.            3292
##  7 2026352035            5567.              5528           2978.             254
##  8 2320127002            4717.              5057           2255.             772
##  9 2347167796            9520.              9781           4682.              42
## 10 2873212765            7556.              7762           1514.            2524
## # ℹ 23 more rows
## # ℹ 2 more variables: max_total_steps <int>, iqr_total_steps <dbl>
# Summary statistics for "TotalSteps" (grouped by type of day)
fitness_data_df %>% 
  group_by(IsWeekend) %>% 
  summarize(
    mean_total_steps = mean(TotalSteps),
    median_total_steps = median(TotalSteps), 
    sd_total_steps = sd(TotalSteps),
    min_total_steps = min(TotalSteps),
    max_total_steps = max(TotalSteps),
    iqr_total_steps = IQR(TotalSteps)
  )
## # A tibble: 2 × 7
##   IsWeekend mean_total_steps median_total_steps sd_total_steps min_total_steps
##   <lgl>                <dbl>              <int>          <dbl>           <int>
## 1 FALSE                7669.               7802          4807.               0
## 2 TRUE                 7551.               6708          5818.               0
## # ℹ 2 more variables: max_total_steps <int>, iqr_total_steps <dbl>

Grouping by ID can be useful for gleaning participant level insights; however, because key personal attributes (like age) were excluded from the dataset, it is more informative to aggregate data by day type (weekday vs. weekend).

When aggregated in this fashion, average step counts are slightly higher on weekdays than on weekdays. This could be due to more structured daily routines, such as commuting. In contrast, the standard deviation of step counts is higher on weekends, indicating greater variability in activity levels and suggestion that individuals’ weekend step counts are less predictable and structured.

ActiveDistance Analysis

There are four categories of activity based on distanced traveled, ranked from highest to lowest intensity: Very Active, Moderately Active, Light Active, and Sedentary Active.

# Summary statistics for "VeryActiveDistance" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_very_active_distance = mean(VeryActiveDistance),
    median_very_active_distance = median(VeryActiveDistance), 
    sd_very_active_distance = sd(VeryActiveDistance),
    min_very_active_distance = min(VeryActiveDistance),
    max_very_active_distance = max(VeryActiveDistance),
    iqr_very_active_distance = IQR(VeryActiveDistance)
  )

# Summary statistics for "ModeratelyActiveDistance" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_moderately_active_distance = mean(ModeratelyActiveDistance),
    median_moderately_active_distance = median(ModeratelyActiveDistance), 
    sd_moderately_active_distance = sd(ModeratelyActiveDistance),
    min_moderately_active_distance = min(ModeratelyActiveDistance),
    max_moderately_active_distance = max(ModeratelyActiveDistance),
    iqr_moderately_active_distance = IQR(ModeratelyActiveDistance)
  )

# Summary statistics for "LightActiveDistance" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_light_active_distance = mean(LightActiveDistance),
    median_light_active_distance = median(LightActiveDistance), 
    sd_light_active_distance = sd(LightActiveDistance),
    min_light_active_distance = min(LightActiveDistance),
    max_light_active_distance = max(LightActiveDistance),
    iqr_light_active_distance = IQR(LightActiveDistance)
  )

# Summary statistics for "SedentaryActiveDistance" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_sedentary_active_distance = mean(SedentaryActiveDistance),
    median_sedentary_active_distance = median(SedentaryActiveDistance), 
    sd_sedentary_active_distance = sd(SedentaryActiveDistance),
    min_sedentary_active_distance = min(SedentaryActiveDistance),
    max_sedentary_active_distance = max(SedentaryActiveDistance),
    iqr_sedentary_active_distance = IQR(SedentaryActiveDistance)
  )
# Average distance traveled by activity level
fitness_data_df %>% 
  summarize(
    mean_very_active_distance = mean(VeryActiveDistance),
    mean_moderately_active_distance = mean(ModeratelyActiveDistance),
    mean_light_active_distance = mean(LightActiveDistance),
    mean_sedentary_active_distance = mean(SedentaryActiveDistance),
  )
##   mean_very_active_distance mean_moderately_active_distance
## 1                  1.502681                       0.5675426
##   mean_light_active_distance mean_sedentary_active_distance
## 1                   3.340819                    0.001606383

The greatest distance traveled on average corresponded to Light Active, followed by Very Active, Moderately Active, and Sedentary Active. This pattern is reasonable because most daily movement transpires at light intensity, with high-intensity movement tending to be shorter and less frequently. Consequently, the total distance accumulated in the Light Active category can exceed that of the Very Active category. The Sedentary Active category shows extremely low distances, reflecting almost no movement.

ActiveMinutes Analysis

There four categories of activity based on minutes active: Very Active, Fairly Active, Lightly Active, and Sedentary Active.

# Summary statistics for "VeryActiveMinutes" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_very_active_minutes = mean(VeryActiveMinutes),
    median_very_active_minutes= median(VeryActiveMinutes), 
    sd_very_active_minutes = sd(VeryActiveMinutes),
    min_very_active_minutes = min(VeryActiveMinutes),
    max_very_active_minutes = max(VeryActiveMinutes),
    iqr_very_active_minutes = IQR(VeryActiveMinutes)
  )

# Summary statistics for "FairlyActiveMinutes" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_fairly_active_minutes = mean(FairlyActiveMinutes),
    median_fairly_active_minutes = median(FairlyActiveMinutes), 
    sd_fairly_active_minutes = sd(FairlyActiveMinutes),
    min_fairly_active_minutes = min(FairlyActiveMinutes),
    max_fairly_active_minutes = max(FairlyActiveMinutes),
    iqr_fairly_active_minutes = IQR(FairlyActiveMinutes)
  )

# Summary statistics for "LightlyActiveMinutes" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_lightly_active_minutes = mean(LightlyActiveMinutes),
    median_lightly_active_minutes = median(LightlyActiveMinutes), 
    sd_lightly_active_minutes = sd(LightlyActiveMinutes),
    min_lightly_active_minutes = min(LightlyActiveMinutes),
    max_lightly_active_minutes = max(LightlyActiveMinutes),
    iqr_lightly_active_minutes = IQR(LightlyActiveMinutes)
  )

# Summary statistics for "SedentaryMinutes" 
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_sedentary_minutes = mean(SedentaryMinutes),
    median_sedentary_minutes = median(SedentaryMinutes), 
    sd_sedentary_minutes = sd(SedentaryMinutes),
    min_sedentary_minutes = min(SedentaryMinutes),
    max_sedentary_minutes = max(SedentaryMinutes),
    iqr_sedentary_minutes = IQR(SedentaryMinutes)
  )
# Average minutes active by activity level
fitness_data_df %>% 
  summarize(
    mean_very_active_minutes = mean(VeryActiveMinutes),
    mean_fairly_active_minutes = mean(FairlyActiveMinutes),
    mean_lightly_active_minutes = mean(LightlyActiveMinutes),
    mean_sedentary_minutes = mean(SedentaryMinutes),
  )
##   mean_very_active_minutes mean_fairly_active_minutes
## 1                 21.16489                   13.56489
##   mean_lightly_active_minutes mean_sedentary_minutes
## 1                    192.8128               991.2106

On average, the most minutes were spent in the Sedentary Active category, followed by Lightly Active, Very Active, and Fairly Active. This pattern is expected, as most of the day is spent inactive. Albeit, the large gap between Sedentary Active and Lightly Active minutes is striking: on average, a total of 991 minutes–or approximately 16 hours and 31 minutes–was spent essentially inactive.

As predicted, Light Active minutes rank second, but participants spent less than 30 minutes on average in Very Active and Fairly Active categories. These results provide an eye-opening insight into participants’ daily activity patterns, showing that most of the day is sedentary with only limited high-intensity activity.

mean_daily_total_steps_df <- fitness_data_df %>% 
  group_by(ActivityDate_Fixed, IsWeekend) %>% 
  summarize(mean_daily_total_steps = mean(TotalSteps))


head(mean_daily_total_steps_df)
glimpse(mean_daily_total_steps_df)

mean_daily_calories_df <- fitness_data_df %>% 
  group_by(ActivityDate_Fixed) %>% 
  summarize(mean_daily_calories = mean(Calories))


head(mean_daily_calories_df)
glimpse(mean_daily_calories_df)

mean_daily_total_steps_calories_df <- merge(mean_daily_total_steps_df, mean_daily_calories_df, by="ActivityDate_Fixed")
head(mean_daily_total_steps_calories_df)
glimpse(mean_daily_total_steps_calories_df)

Correlation between Average Daily Calories and Avery Daily Steps

In statistics, the Pearson correlation coefficient (r) is a number between -1 and 1 that quantifies the strength and direction of a linear relationship between two variables. Values of r close to 1 indicate a strong positive relationship, values close to -1 indicate a strong negative relationship, and values close to 0 represent little to no linear correlation.

# Calculating the correlation coefficient between MeanDailyTotalSteps and MeanDailyTotalCalories 
r_value_calories_steps <- cor(mean_daily_total_steps_calories_df$mean_daily_total_steps, mean_daily_total_steps_calories_df$mean_daily_calories, use="complete.obs")

r_value_calories_steps
## [1] 0.895497

The Pearson correlation coefficient between the average daily steps and average daily calories burned is approximately 0.9, which is indicative of a very strong positive linear relationship. In other words, as steps increase, calories burned tend to increase.

Correlation Matrix for the Daily Activity Dataset

Next, the relationships between key numeric variables in the Daily Activity dataset were analyzed via a correlation matrix.

activity_correlation_matrix_df <- fitness_data_df[, c("TotalSteps", "TotalDistance", "VeryActiveMinutes", "FairlyActiveMinutes", "LightlyActiveMinutes", "SedentaryMinutes", "Calories")]

activity_correlation_matrix <- round(cor(activity_correlation_matrix_df), 2)
print(activity_correlation_matrix)
##                      TotalSteps TotalDistance VeryActiveMinutes
## TotalSteps                 1.00          0.99              0.67
## TotalDistance              0.99          1.00              0.68
## VeryActiveMinutes          0.67          0.68              1.00
## FairlyActiveMinutes        0.50          0.46              0.31
## LightlyActiveMinutes       0.57          0.52              0.05
## SedentaryMinutes          -0.33         -0.29             -0.16
## Calories                   0.59          0.64              0.62
##                      FairlyActiveMinutes LightlyActiveMinutes SedentaryMinutes
## TotalSteps                          0.50                 0.57            -0.33
## TotalDistance                       0.46                 0.52            -0.29
## VeryActiveMinutes                   0.31                 0.05            -0.16
## FairlyActiveMinutes                 1.00                 0.15            -0.24
## LightlyActiveMinutes                0.15                 1.00            -0.44
## SedentaryMinutes                   -0.24                -0.44             1.00
## Calories                            0.30                 0.29            -0.11
##                      Calories
## TotalSteps               0.59
## TotalDistance            0.64
## VeryActiveMinutes        0.62
## FairlyActiveMinutes      0.30
## LightlyActiveMinutes     0.29
## SedentaryMinutes        -0.11
## Calories                 1.00

The Pearson correlation coefficient between TotalSteps and TotalDistance is about 0.99, which indicates a very strong positive linear relationship. This is to be expected because the more steps someone takes, the greater distance they travel.

The Pearson correlation coefficient between TotalSteps and VeryActiveMinutes is roughly 0.67, which represents a strong positive linear relationship. This demonstrates that people who take more steps tend to have more intense activity.

The Pearson correlation coefficient between TotalSteps and SedentaryMinutes is approximately -0.16, which denotes a weak negative linear relationship. This follows, as days with more steps generally include less sedentary time.

Data Analysis on the Daily Sleep Dataset

Next, the key variables in the Daily Sleep Dataset were analyzed.

TotalMinutesAsleep Analysis

Summary statistics were computed for the TotalMinutesAsleep variable–grouped by participant and type of day, respectively.

# Summary statistics for "TotalMinutesAsleep" (grouped by participant)
fitness_data_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_total_minutes_asleep = mean(TotalMinutesAsleep, na.rm = TRUE),
    median_total_minutes_asleep = median(TotalMinutesAsleep, na.rm = TRUE), 
    sd_total_minutes_asleep = sd(TotalMinutesAsleep, na.rm = TRUE),
    min_total_minutes_asleep = min(TotalMinutesAsleep, na.rm = TRUE),
    max_total_minutes_asleep = max(TotalMinutesAsleep, na.rm = TRUE),
    iqr_total_minutes_asleep = IQR(TotalMinutesAsleep, na.rm = TRUE)
  )

# Summary statistics for "TotalMinutesAsleep" (grouped by type of day)
fitness_data_df %>% 
  group_by(IsWeekend) %>% 
  summarize(
    mean_total_minutes_asleep = mean(TotalMinutesAsleep, na.rm = TRUE),
    median_total_minutes_asleep= median(TotalMinutesAsleep, na.rm = TRUE), 
    sd_total_minutes_asleep = sd(TotalMinutesAsleep, na.rm = TRUE),
    min_total_minutes_asleep = min(TotalMinutesAsleep, na.rm = TRUE),
    max_total_minutes_asleep = max(TotalMinutesAsleep, na.rm = TRUE),
    iqr_total_minutes_asleep = IQR(TotalMinutesAsleep, na.rm = TRUE)
  )

When arranged type of day, average minutes spent asleep are slightly higher on weekends than on weekdays. This could be attributed to people waking up on earlier on weekdays for work or school, fewer obligations on weekends, and biological rhythms which naturally attempt to extend sleep duration on weekends. The standard deviation of step counts is also greater on weekends, indicating greater variability in wake-up times or differences in lifestyle habits.

A New Feature: The SleepEfficiency Column

The Daily Sleep dataset includes TotalMinutesAsleep and TotalTimeInBed columns. While these metrics provide useful insights on their own, calculating sleep efficiency–the ratio of total time asleep to total time in bed–enables a better understanding of sleep quality. It can be calculated using the following equation: (TotalMinutesAsleep / TotalTimeInBed) * 100

Sleep efficiency contextualizes sleep duration by showing how effectively time in bed is used for rest.

fitness_data_sleep_df <- fitness_data_df

# Adding in a new column, SleepEfficiency, the ratio of total time asleep to total time in bed 
fitness_data_sleep_df$SleepEfficiency <- ((fitness_data_sleep_df$TotalMinutesAsleep / fitness_data_sleep_df$TotalTimeInBed) * 100)
fitness_data_sleep_df
fitness_data_sleep_df <- na.omit(fitness_data_sleep_df)
##           Id ActivityDate_Fixed TotalSteps TotalDistance TrackerDistance
## 1 1503960366         2016-04-12      13162          8.50            8.50
## 2 1503960366         2016-04-13      10735          6.97            6.97
## 4 1503960366         2016-04-15       9762          6.28            6.28
## 5 1503960366         2016-04-16      12669          8.16            8.16
## 6 1503960366         2016-04-17       9705          6.48            6.48
## 8 1503960366         2016-04-19      15506          9.88            9.88
##   VeryActiveDistance ModeratelyActiveDistance LightActiveDistance
## 1               1.88                     0.55                6.06
## 2               1.57                     0.69                4.71
## 4               2.14                     1.26                2.83
## 5               2.71                     0.41                5.04
## 6               3.19                     0.78                2.51
## 8               3.53                     1.32                5.03
##   SedentaryActiveDistance VeryActiveMinutes FairlyActiveMinutes
## 1                       0                25                  13
## 2                       0                21                  19
## 4                       0                29                  34
## 5                       0                36                  10
## 6                       0                38                  20
## 8                       0                50                  31
##   LightlyActiveMinutes SedentaryMinutes Calories IsWeekend TotalSleepRecords
## 1                  328              728     1985     FALSE                 1
## 2                  217              776     1797     FALSE                 2
## 4                  209              726     1745     FALSE                 1
## 5                  221              773     1863      TRUE                 2
## 6                  164              539     1728      TRUE                 1
## 8                  264              775     2035     FALSE                 1
##   TotalMinutesAsleep TotalTimeInBed SleepEfficiency
## 1                327            346        94.50867
## 2                384            407        94.34889
## 4                412            442        93.21267
## 5                340            367        92.64305
## 6                700            712        98.31461
## 8                304            320        95.00000
# Summary statistics for "SleepEfficiency" (grouped by participant)
fitness_data_sleep_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_sleep_efficiency = mean(SleepEfficiency, na.rm = TRUE),
    median_sleep_efficiency = median(SleepEfficiency, na.rm = TRUE), 
    sd_sleep_efficiency = sd(SleepEfficiency, na.rm = TRUE),
    min_sleep_efficiency = min(SleepEfficiency, na.rm = TRUE),
    max_sleep_efficiency = max(SleepEfficiency, na.rm = TRUE),
    iqr_sleep_efficiency = IQR(SleepEfficiency, na.rm = TRUE)
  )
## # A tibble: 24 × 7
##            Id mean_sleep_efficiency median_sleep_efficiency sd_sleep_efficiency
##         <dbl>                 <dbl>                   <dbl>               <dbl>
##  1 1503960366                  93.6                    94.0                2.97
##  2 1644430081                  88.2                    88.1                4.49
##  3 1844505072                  67.8                    67.0                6.91
##  4 1927972279                  94.7                    94.3                1.35
##  5 2026352035                  94.1                    94.2                2.13
##  6 2320127002                  88.4                    88.4               NA   
##  7 2347167796                  91.0                    90.5                3.01
##  8 3977333714                  63.4                    63.1                7.05
##  9 4020332650                  93.0                    91.9                3.84
## 10 4319703577                  94.7                    94.8                1.75
## # ℹ 14 more rows
## # ℹ 3 more variables: min_sleep_efficiency <dbl>, max_sleep_efficiency <dbl>,
## #   iqr_sleep_efficiency <dbl>
# Summary statistics for "SleepEfficiency" (grouped by type of day)
fitness_data_sleep_df %>% 
  group_by(IsWeekend) %>% 
  summarize(
    mean_sleep_efficiency = mean(SleepEfficiency, na.rm = TRUE),
    median_sleep_efficiency = median(SleepEfficiency, na.rm = TRUE), 
    sd_sleep_efficiency = sd(SleepEfficiency, na.rm = TRUE),
    min_sleep_efficiency = min(SleepEfficiency, na.rm = TRUE),
    max_sleep_efficiency = max(SleepEfficiency, na.rm = TRUE),
    iqr_sleep_efficiency = IQR(SleepEfficiency, na.rm = TRUE)
  )
## # A tibble: 2 × 7
##   IsWeekend mean_sleep_efficiency median_sleep_efficiency sd_sleep_efficiency
##   <lgl>                     <dbl>                   <dbl>               <dbl>
## 1 FALSE                      91.9                    94.3                8.22
## 2 TRUE                       91.0                    94.3                9.98
## # ℹ 3 more variables: min_sleep_efficiency <dbl>, max_sleep_efficiency <dbl>,
## #   iqr_sleep_efficiency <dbl>

Most participants have a sleep efficiency in the high 80s to high 90s percentile, with the exception of two individuals, whose sleep efficiency is notably lower at approximately 67.8% and 63.4%, respectively. When grouped by type of day, average sleep efficiency is quite uniform between weekdays and weekends, with weekdays showing a slightly higher value–both averaging around 91%. The standard deviation figures are also similar, indicating that participants overall maintain a good and consistent sleep quality.

sleep_correlation_matrix_df <- fitness_data_sleep_df[, c("TotalMinutesAsleep", "TotalTimeInBed", "SleepEfficiency")]

# Calculating the correlation coefficient between "TotalMinutesAsleep" and "TotalTimeInBed" 
r_value_minutes_asleep_bed <- cor(fitness_data_sleep_df$TotalMinutesAsleep, fitness_data_sleep_df$TotalTimeInBed, use="complete.obs")
r_value_minutes_asleep_bed
## [1] 0.9304224
# Calculating the correlation coefficient between "TotalMinutesAsleep" and "SleepEfficiency" 
r_value_minutes_alseep__sleep_efficiency <- cor(fitness_data_sleep_df$TotalMinutesAsleep, fitness_data_sleep_df$SleepEfficiency, use="complete.obs")
r_value_minutes_alseep__sleep_efficiency
## [1] 0.2645268
# Calculating the correlation coefficient between "TotalTimeInBed" and "SleepEfficiency" 
r_value_bed_sleep_efficiency <- cor(fitness_data_sleep_df$TotalTimeInBed, fitness_data_sleep_df$SleepEfficiency, use="complete.obs") 
r_value_bed_sleep_efficiency
## [1] -0.09111555
sleep_correlation_matrix <- round(cor(sleep_correlation_matrix_df), 2)
print(sleep_correlation_matrix)
##                    TotalMinutesAsleep TotalTimeInBed SleepEfficiency
## TotalMinutesAsleep               1.00           0.93            0.26
## TotalTimeInBed                   0.93           1.00           -0.09
## SleepEfficiency                  0.26          -0.09            1.00

The Pearson correlation coefficient between TotalMinutesAsleep and TotalTimeInBed is approximately 0.93, which indicates a very strong positive linear relationship. Therefore, as time in bed increases, total minutes asleep also increase.

The Pearson correlation coefficient between TotalMinutesAsleep and SleepEfficiency is roughly 0.27, which shows a weak positive linear relationship. This suggests that longer sleep duration is only slightly associated with higher sleep efficiency.

The Pearson correlation coefficient between TotalTimeInBed and SleepEfficiency is approximately -0.09, which signifies a very weak negative linear relationship. This implies that there is little to no relationship between time spent in bed and sleep quality.

Data Analysis on the Joint Daily Activity and Daily Sleep Datasets

Finally, an integrated data analysis was conducted on the merged Daily Activity and Daily Sleep datasets to pinpoint relationships between physical activity and sleep metrics.

A New Feature: The StepBand Column

To enable a deeper analysis of the relationship between Daily Activity and Daily Sleep, a new column StepBand was added to the dataframe. Each participant is assigned to one of three step bands–Low, Medium, or High–based on whether their TotalSteps count falls below the first quartile, between the first and third quartiles, or above the third quartile of the TotalSteps distribution.

mean_daily_sleep_efficiency_df <- fitness_data_sleep_df %>% 
  group_by(ActivityDate_Fixed, IsWeekend) %>% 
  summarize(mean_daily_sleep_efficiency = mean(SleepEfficiency))


head(mean_daily_sleep_efficiency_df)
glimpse(mean_daily_sleep_efficiency_df)

fitness_data_sleep_df_2 <- fitness_data_sleep_df %>% 
  select("TotalSteps", "TotalMinutesAsleep", "SleepEfficiency", "IsWeekend")

head(fitness_data_sleep_df_2)

steps_list<- fitness_data_sleep_df_2$TotalSteps

# Determining the first and third quartiles of TotalSteps to better understand the spread of the data 
q1_steps <- quantile(fitness_data_sleep_df_2$TotalSteps, 0.25)
q3_steps <- quantile(fitness_data_sleep_df_2$TotalSteps, 0.75)


# Adding in a new column, StepBand, which categorizes total steps as Low, Medium, or High relative to the first and third quartiles 
fitness_data_step_band_df <- fitness_data_sleep_df_2 %>% 
  mutate(StepBand = case_when (
    TotalSteps < q1_steps ~ 'Low',
    TotalSteps > q3_steps ~ 'High',
    TRUE ~ 'Medium'
  ))
##   TotalSteps TotalMinutesAsleep SleepEfficiency IsWeekend StepBand
## 1      13162                327        94.50867     FALSE     High
## 2      10735                384        94.34889     FALSE   Medium
## 4       9762                412        93.21267     FALSE   Medium
## 5      12669                340        92.64305      TRUE     High
## 6       9705                700        98.31461      TRUE   Medium
## 8      15506                304        95.00000     FALSE     High

Correlation between Total Steps, Minutes Asleep, and Sleep Efficiency

Relationships between key variables across both domains were determined via a correlation matrix.

# Calculating the correlation coefficient between "TotalSteps" and "TotalMinutesAsleep" 
r_value_steps_minutes_asleep <- cor(fitness_data_sleep_df$TotalSteps, fitness_data_sleep_df$TotalMinutesAsleep, use="complete.obs")
r_value_steps_minutes_asleep
## [1] -0.1903439
# Calculating the correlation coefficient between "TotalSteps" and "SleepEfficiency" 
r_value_steps_sleep_efficiency <- cor(fitness_data_sleep_df$TotalSteps, fitness_data_sleep_df$SleepEfficiency, use="complete.obs")
r_value_steps_sleep_efficiency
## [1] -0.1100255
# Calculating the correlation coefficient between "Calories" and "SleepEfficiency" 
r_value_calories_sleep_efficiency <- cor(fitness_data_sleep_df$Calories, fitness_data_sleep_df$SleepEfficiency, use="complete.obs") 
r_value_calories_sleep_efficiency
## [1] 0.2948618
activity_sleep_correlation_matrix_df <- fitness_data_sleep_df [, c("TotalSteps", "TotalDistance", "Calories", "TotalMinutesAsleep", "SleepEfficiency")]
clean_activity_sleep_correlation_matrix_df <- activity_sleep_correlation_matrix_df  %>%
  drop_na()
clean_activity_sleep_correlation_matrix <- round(cor(clean_activity_sleep_correlation_matrix_df), 2)
print(clean_activity_sleep_correlation_matrix)
##                    TotalSteps TotalDistance Calories TotalMinutesAsleep
## TotalSteps               1.00          0.98     0.41              -0.19
## TotalDistance            0.98          1.00     0.52              -0.18
## Calories                 0.41          0.52     1.00              -0.03
## TotalMinutesAsleep      -0.19         -0.18    -0.03               1.00
## SleepEfficiency         -0.11         -0.08     0.29               0.26
##                    SleepEfficiency
## TotalSteps                   -0.11
## TotalDistance                -0.08
## Calories                      0.29
## TotalMinutesAsleep            0.26
## SleepEfficiency               1.00

The Pearson correlation coefficient between the total daily steps and total minutes asleep is approximately -0.19, which indicates a weak negative linear relationship. This suggests that there is only a slight tendency for higher activity to be associated with shorter sleep duration, albeit the relationship is minimal.

The Pearson correlation coefficient between the total daily steps and sleep efficiency is roughly -0.11, which represents a very weak negative linear relationship. As such, there is virtually no association between activity level and sleep quality.

The Pearson correlation coefficient between calories burned and sleep efficiency about 0.29, which signifies a weak positive linear relationship. This demonstrates a modest tendency for higher sleep quality to be associated with greater energy expenditure.

A New Feature: The BehavioralProfile Column

In order to gauge the distribution of participants using the available activity and sleep data, new BehavioralProfile column was added. Similar to the previously introduced StepBand column, this feature assigns participants to one of four behavioral profiles: Lazy Sunday, Weekend Warrior, Inactive but Sleep Loss, and Active but Sleep Loss.

Profiles are determined according to changes in steps and sleep between weekdays and weekends:

Behavioral profiles are assigned as follows:

# Calculating the change in steps and minutes asleep using the mean steps and minutes asleep on weekdays and weekends, respectively
fitness_data_sleep_mean_steps_sleep <- fitness_data_sleep_df %>% 
  group_by(Id) %>% 
  summarize(
    mean_weekday_steps = mean(TotalSteps[!IsWeekend], na.rm=TRUE),
    mean_weekend_steps = mean(TotalSteps[IsWeekend], na.rm=TRUE), 
    mean_weekday_sleep = mean(TotalMinutesAsleep[!IsWeekend], na.rm=TRUE),
    mean_weekend_sleep = mean(TotalMinutesAsleep[IsWeekend], na.rm=TRUE)
  ) %>% 
mutate(
  delta_steps = mean_weekend_steps - mean_weekday_steps,
  delta_sleep = mean_weekend_sleep - mean_weekday_sleep
)


fitness_data_sleep_mean_steps_sleep <- fitness_data_sleep_mean_steps_sleep %>% 
  filter(!is.na(delta_steps))

# Adding in a new column, BehavioralProfile, which determines participants' behavioral profile--Lazy Sunday, Weekend Warrior, Inactive / Poor Sleep, or Active but Sleep Loss--based on their DeltaSteps and DeltaSleep values
fitness_data_sleep_mean_steps_sleep <- fitness_data_sleep_mean_steps_sleep %>% 
  mutate(BehavioralProfile = case_when (
    delta_steps < 0 & delta_sleep > 0 ~ 'Lazy Sunday',
    delta_steps > 0 & delta_sleep > 0 ~ 'Weekend Warrior',
    delta_steps < 0 & delta_sleep < 0 ~ 'Inactive / Poor Sleep',
    TRUE ~ 'Active but Sleep Loss'
  ))
## # A tibble: 6 × 4
##           Id delta_steps delta_sleep BehavioralProfile    
##        <dbl>       <dbl>       <dbl> <chr>                
## 1 1503960366       -944.      119.   Lazy Sunday          
## 2 1644430081       9002.     -327    Active but Sleep Loss
## 3 1844505072       -550.       12    Lazy Sunday          
## 4 2026352035      -1340.       24.6  Lazy Sunday          
## 5 2347167796       2602.       13.2  Weekend Warrior      
## 6 3977333714       1590.       -4.23 Active but Sleep Loss
## # A tibble: 20 × 4
##            Id delta_steps delta_sleep BehavioralProfile    
##         <dbl>       <dbl>       <dbl> <chr>                
##  1 1503960366       -944.      119.   Lazy Sunday          
##  2 1644430081       9002.     -327    Active but Sleep Loss
##  3 1844505072       -550.       12    Lazy Sunday          
##  4 2026352035      -1340.       24.6  Lazy Sunday          
##  5 2347167796       2602.       13.2  Weekend Warrior      
##  6 3977333714       1590.       -4.23 Active but Sleep Loss
##  7 4020332650      -3566.     -172.   Inactive / Poor Sleep
##  8 4319703577      -2940.       80.4  Lazy Sunday          
##  9 4388161847       1783.      159.   Weekend Warrior      
## 10 4445114986        343.      -48.4  Active but Sleep Loss
## 11 4558609924      -5256.      -14.3  Inactive / Poor Sleep
## 12 4702921684       5209.       88.3  Weekend Warrior      
## 13 5553957443      -7116.      149.   Lazy Sunday          
## 14 5577150313       4349.       17.6  Weekend Warrior      
## 15 6117666160        591.      -53.9  Active but Sleep Loss
## 16 6962181067       -573.      -15.0  Inactive / Poor Sleep
## 17 7086361926        637.       75.6  Weekend Warrior      
## 18 8053475328       5956      -284.   Active but Sleep Loss
## 19 8378563200      -4248.       91.8  Lazy Sunday          
## 20 8792009665       1545.      -50.8  Active but Sleep Loss
# Calculating the number of participants in each behavioral profile
count_group_df <- fitness_data_sleep_mean_steps_sleep %>% 
  group_by(BehavioralProfile) %>% 
  summarize(count_group = n()
  )

count_group_df
## # A tibble: 4 × 2
##   BehavioralProfile     count_group
##   <chr>                       <int>
## 1 Active but Sleep Loss           6
## 2 Inactive / Poor Sleep           3
## 3 Lazy Sunday                     6
## 4 Weekend Warrior                 5

The distribution of behavioral profiles indicates that Lazy Sunday and Active but Sleep* are the most common, each with 6 participants (30%), followed by Weekend Warrior with 5 participants (25%), and Inactive / Poor Sleep with 3 participants (15%).

Participants in the Lazy Sunday group have lesser activity but relatively more sleep, while those in the Active but Sleep Loss group are physically active but sleep less. Weekend Warriors attain a balance of activity and sleep, whereas Inactive / Poor Sleep participants show notable deficits in both.

On balance, the fairly even distribution highlights variability in participant behavior, with most individuals falling into unbalanced profiles, suggesting opportunities to improve either activity, sleep, or both.

VI. The Share Phase

During the Share phase, data visualizations are created based on key insights to effectively communicate findings.

Creating Visualizations for the Daily Activity Dataset

To start, visualizations highlighting the key features and dimensions of the Daily Activity dataset were created.

Step Count By Participant Boxplot

To start, a boxplot showing step count by participant was generated. Boxplots are effective for comparing multiple groups, identifying outliers, and getting a sense of the overall distribution. Assuming that this dataset consists of adults, the minimum recommended step count is 10,000 steps. This boxplot displays which participants are meeting, exceeding, or falling below this recommendation.

Overall Step Count During Weekdays and Weekends Boxplot

Next, a boxplot showing the overall step count, stratified by type of day, was produced. This visualization indicates that participants take slightly more steps on weekdays compared to weekends on average. However, the highest recorded step count appears as an outlier and transpired on a weekend.

Overall Distance Traveled During Weekdays and Weekends Boxplot

Similarly, a boxplot showing the overall distance traveled, stratified by type of day, was generated. As expected, this visualization closely mirrors the step count distribution, as distance traveled and step count are directly related metrics.

Average Distance Traveled By Activity Level Bar Chart

Next, a bar chart depicting the average distance traveled, stratified by activity level, was created. The Lightly Active group recorded the highest average distance traveled, followed by the Very Active, Moderately Active, and Sedentary Active groups.

##   VeryActiveDistance ModeratelyActiveDistance LightActiveDistance
## 1               1.88                     0.55                6.06
## 2               1.57                     0.69                4.71
## 3               2.44                     0.40                3.91
## 4               2.14                     1.26                2.83
## 5               2.71                     0.41                5.04
## 6               3.19                     0.78                2.51
##   SedentaryActiveDistance IsWeekend
## 1                       0     FALSE
## 2                       0     FALSE
## 3                       0     FALSE
## 4                       0     FALSE
## 5                       0      TRUE
## 6                       0      TRUE
## Rows: 940
## Columns: 5
## $ VeryActiveDistance       <dbl> 1.88, 1.57, 2.44, 2.14, 2.71, 3.19, 3.25, 3.5…
## $ ModeratelyActiveDistance <dbl> 0.55, 0.69, 0.40, 1.26, 0.41, 0.78, 0.64, 1.3…
## $ LightActiveDistance      <dbl> 6.06, 4.71, 3.91, 2.83, 5.04, 2.51, 4.71, 5.0…
## $ SedentaryActiveDistance  <dbl> 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, …
## $ IsWeekend                <lgl> FALSE, FALSE, FALSE, FALSE, TRUE, TRUE, FALSE…
## # A tibble: 6 × 3
##   IsWeekend DistanceType             Distance
##   <lgl>     <chr>                       <dbl>
## 1 FALSE     VeryActiveDistance          1.88 
## 2 FALSE     ModeratelyActiveDistance    0.550
## 3 FALSE     LightActiveDistance         6.06 
## 4 FALSE     SedentaryActiveDistance     0    
## 5 FALSE     VeryActiveDistance          1.57 
## 6 FALSE     ModeratelyActiveDistance    0.690
## Rows: 3,760
## Columns: 3
## $ IsWeekend    <lgl> FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, FALSE, F…
## $ DistanceType <chr> "VeryActiveDistance", "ModeratelyActiveDistance", "LightA…
## $ Distance     <dbl> 1.88, 0.55, 6.06, 0.00, 1.57, 0.69, 4.71, 0.00, 2.44, 0.4…

Average Minutes Active By Activity Level Bar Chart

Likewise, a bar chart depicting the average minutes active, stratified by activity level, was generated. The Sedentary Active group recorded the highest average minutes by a substantial margin, followed by the Moderately Active, Very Active, and Lightly Active groups.

Average Daily Step Count Over One Month Line Chart

Next, a line chart showcasing the daily step count, stratified by type of day–with red and blue lines representing weekdays and weekends, respectively–was generated. Since the data spans a period of time, a line chart is the most appropriate visualization to use. The weekend trend shows a marked decline in step counts, with the lowest values occurring on Sundays–supporting the Lazy Sunday behavioral profile.

Average Daily Calories Burned by Average Daily Total Steps Scatterplot

Subsequently, a scatterplot chart highlighting the relationship between average daily calories burned and average daily total steps, stratified by type of day, was created. Since the goal was to evaluate the correlation between two continuous variables, a scatterplot is the most appropriate visualization. A linear regression line was also drawn to represent the line of best fit. Points that lie close to the line denote a strong linear relationship, whereas father from the line indicate greater variability and a weaker association between the variables.

Correlation Matrix for the Daily Activity Dataset

Finally, a correlation matrix, which examines the relationship between the variables in the Daily Activity dataset, was generated. In the matrix, darker colors represent strong correlations, while lighter colors indicate weaker correlations.

Creating Visualizations for the Daily Sleep Dataset

Next, visualizations highlighting the key features and dimensions of the Daily Sleep dataset were produced.

Minutes Asleep By Participant Boxplot

First, a boxplot depicting the minutes asleep by participant was produced. Assuming that this dataset consists of adults, the minimum recommended sleep duration is 420 minutes (7 hours). This boxplot shows which participants are meeting, exceeding, or falling below this recommendation.

Minutes Asleep During Weekdays and Weekends Boxplot

Next, a boxplot showing the overall minutes spent asleep, stratified by type of day, was produced. This visualization indicates that participants, on average, get slightly more sleep on weekends than on weekdays. However, the highest sleep duration appears as an outlier and transpired on a weekday.

Average Daily Minutes Asleep Over One Month Line Chart

Next, a line chart depicting the average daily minutes asleep over one month, stratified by type of day, was created. The weekend trend shows a considerable peak–particularly on Sundays–where the highest values on the chart occur, further reinforcing the Lazy Sunday behavioral profile.

Overall Sleep Efficiency During Weekdays and Weekends Boxplot

Then, a boxplot displaying overall sleep efficiency, stratified by type of day, was generated. Sleep quality appears relatively consistent irrespective of whether it it is a weekday or weekend, though it is slightly higher on weekdays.

Correlation Matrix for the Daily Sleep Dataset

Finally, a correlation matrix, which surveys the relationship between the variables in the Daily Sleep dataset, was created.

Creating Visualizations for the Joint Daily Activity and Daily Sleep Datasets

To conclude the Share phase, visualizations highlighting the relationships between key variables in the merged Daily Activity and Daily Sleep datasets were created.

Minutes Asleep By Sleep Efficiency Scatterplot

First, a scatterplot chart highlighting the relationship between total minutes asleep and sleep efficiency, stratified by step band, was created. While grouping the data in this way links step count to sleep quality, step band does not appear to be a significant driver of sleep efficiency. Nevertheless, the chart suggests that sleep efficiency tends to improve as total sleep duration increases.

Correlation Matrix for the Merged Daily Activity and Daily Sleep Datasets

Next, a correlation matrix, which analyzes the relationship between the variables across both datasets, was generated.

Minutes Asleep by Sleep Efficiency Scatterplot

Next, a scatterplot highlighting the relationship between daily total steps and daily sleep efficiency, stratified by participant, was produced. Even though step count varies significantly across participants it has negligible relationship with sleep quality, indicating that higher physical activities levels do not innately result in improved sleep efficiency.

Percentage Distribution of Participants’ Activity-Sleep Behavioral Profiles Bar Chart

Subsequently, a bar chart displaying the percentage distribution of participants’ activity-sleep behavioral profiles was created. These behavioral profiles unite step count and minutes asleep into a single composite behavioral classification. To this end, the most common behavioral profiles pinpointed are Lazy Sunday and Active but Sleep Loss.

Change in Minutes Asleep By Change in Total Steps Scatterplot

Finally, a scatterplot displaying the change in minutes asleep relative to the change in total steps for each participant, stratified by behavioral profile, was produced. Even though it is technically a scatterplot, the chart resembles a Cartesian plane, making it easy to identify the quadrant in which each participant falls. Simply put, each individual in the study is represented by a dot, with the color corresponding to their behavioral profile.

VII. The Act Phase

During the Act phase, the obtained insights are applied to inform recommendations and business decisions.

Recommendation 1: Implement a Reminder System to Encourage Weekend Activity

The analysis uncovered that users are typically less active and sleep more on Sundays. While rest is important, encouraging light physical activity on weekends can help maintain overall health. Therefore, it is recommended that the Leaf app utilizes users’ weekend activity data to send time, gentle reminders throughout the day. These notifications would endorse movement–such as short walks, stretching, or light exercise–while still respecting users’ rest periods. This feature strives to increase Leaf usage and weekend engagement, as well as support healthier, more balanced activity patterns.

Recommendation 2: Implement Personalized Sleep Quality Guidance

The analysis revealed that sleep efficiency improves with longer sleep, although some participants exhibited irregular sleep patterns. To address this, it is recommended that the Leaf app provides personalized notifications, such as bedtime reminders, guided relaxation exercises, or activity-based sleep suggestions. These features would promote healthier sleep habits, enhance overall sleep quality, strengthen user trust in Leaf’s data driven guidance, and potentially increase adoption of premium features.

Recommendation 3: Implement a Behavioral Profile System

One of the crucial features developed for this analysis was the behavioral profile classification, which categorized participants based on changes in minutes asleep relative to changes in total steps. To leverage these insights, it is recommended that the Leaf app implement a behavioral profile system that classifies users according to their activity-sleep patterns and customizes the user experience accordingly. For example, users identified as Lazy Sunday could receive notifications encouraging light weekend activity, while those classified as Active but Sleep Loss could receive sleep-focused guidance and recovery recommendations. By delivering personalized and behavior-based insights, the company can enhance user engagement, improve retention, and extend the perceived value of the Leaf wearable.

VIII. Conclusion

Over the course of the Bellabeat Case Study project, health data–including activity and sleep metrics–was cleaned, processed were cleaned, processed, analyzed, and visualized. This analysis uncovered meaningful patterns and relationships within and across datasets, providing actionable insights into user behavior. These insights informed three business recommendations designed to improve user engagement, support healthier habits, and increase the overall value of the Leaf device. Bu implementing these recommendations, Bellabeat can harness data-driven strategies to strengthen user experience, prompt long-term retention, and reinforce its position as a leader in women’s wellness technology.